There has been conventionally proposed an image process for separating a single image into a plurality of regions. According to one example of such an image process, a suitable threshold is first used to convert an entire image into a binary image. Character rectangles are found out by extracting groups of connected black pixels out of the binary image. A threshold value is determined for each rectangle, and each rectangle is locally converted into a binary image by using the determined threshold value. That is, each rectangle is locally binarized by using the determined threshold value. A threshold value for the entire image is then determined based on the threshold value for each rectangle. Parts of the entire image other than the already-binarized portions are binarized by using the threshold value determined for the entire image. This method is described in Japanese Patent Application Publication No. 2001-291058, for example.